add profiling using cprofile

.circleci/config.yml

@@ -329,7 +329,7 @@ jobs:
             coverage run flood_forecast/trainer.py -p tests/custom_encode.json
             coverage run flood_forecast/trainer.py -p tests/multi_decoder_test.json
             coverage run flood_forecast/trainer.py -p tests/test_dual.json
-            coverage run flood_forecast/trainer.py -p tests/dsanet.json
+            python -m cProfile -s cumulative flood_forecast/trainer.py -p tests/dsanet.json
 
       - store_test_results:
           path: test-results

flood_forecast/basic/gru_vanilla.py

@@ -9,7 +9,7 @@ def __init__(self, n_time_series: int, hidden_dim: int, num_layers: int, n_targe
 
         :param n_time_series: The number of time series present in the data
         :type n_time_series int:
-        :param hidden_dim:
+        :param hidden_dim: The hidden dimension of the GRU
         :type hidden_dim:
 
         Note for probablistic n_targets must be set to two and actual multiple targs are not supported now.

flood_forecast/preprocessing/closest_station.py

@@ -14,7 +14,7 @@ def get_closest_gage(
         start_row: int,
         end_row: int):
     # Function that calculates the closest weather stations to gage and stores in JSON
-    # Base u
+    # Base u r
     for row in range(start_row, end_row):
         gage_info = {}
         gage_info["river_id"] = int(gage_df.iloc[row]['id'])

flood_forecast/pytorch_training.py

@@ -12,6 +12,9 @@
 from flood_forecast.training_utils import EarlyStopper
 from flood_forecast.custom.custom_opt import GaussianLoss, MASELoss
 from torch.nn import CrossEntropyLoss
+import line_profiler
+
+profile = line_profiler.LineProfiler()
 
 
 def multi_crit(crit_multi: List, output, labels, valid=None):
@@ -220,6 +223,7 @@ def train_transformer_style(
     print("test loss:", test)
     model.params["run"] = session_params
     model.save_model(model_filepath, max_epochs)
+    profile.print_stats()
 
 
 def get_meta_representation(column_id: str, uuid: str, meta_model: PyTorchForecast) -> torch.Tensor:
@@ -340,6 +344,7 @@ def compute_loss(labels, output, src, criterion, validation_dataset, probabilist
     return loss
 
 
+@profile
 def torch_single_train(model: PyTorchForecast,
                        opt: optim.Optimizer,
                        criterion: Type[torch.nn.modules.loss._Loss],

requirements.txt

@@ -21,3 +21,4 @@ sphinx
 sphinx-rtd-theme
 sphinx-autodoc-typehints
 sphinx
+line_profiler

setup.py

@@ -16,7 +16,7 @@
 
 setup(
     name='flood_forecast',
-    version='1.0000dev',
+    version='1.0dev',
     packages=[
         'flood_forecast',
         'flood_forecast.transformer_xl',